System, method and apparatus for training a machine learning model

Abstract

The disclosed systems, structures, and methods are directed to receiving a training data set comprising a plurality of original training samples, augmenting the original training samples by applying default transformations, training the machine learning model on at least a portion of the original training samples and at least a portion of the first set of augmented training samples, computing an unaugmented accuracy, augmenting the original training samples and the first set of augmented training samples by applying a candidate transformation, training the machine learning model on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, and at least a portion of the second set of augmented training samples, computing an augmented accuracy, computing an affinity metric from the unaugmented accuracy and the augmented accuracy, and updating the candidate augmentation transformations list and the default augmentation transformations list.

What is claimed is: 
     
       1. A method for training a machine learning model comprising:
 receiving a training data set for training the machine learning model to perform a machine learning task, the training data set comprising a plurality of original training samples; 
 augmenting the original training samples by applying default transformations stored in a default augmentation transformations list and storing the augmented training samples as a first set of augmented training samples in the training data set; 
 training the machine learning model on at least a portion of the original training samples and at least a portion of the first set of augmented training samples; 
 computing an unaugmented accuracy of the machine learning model trained on at least a portion of the original training samples and at least a portion of the first set of augmented training samples; 
 augmenting the original training samples and the first set of augmented training samples by applying a candidate transformation selected from a candidate augmentation transformations list and storing the augmented training samples as a second set of augmented training samples in the training data set; 
 training the machine learning model on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, and at least a portion of the second set of augmented training samples; 
 computing an augmented accuracy of the machine learning model trained on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, at least a portion of the second set of augmented training samples; 
 computing an affinity metric from the unaugmented accuracy and the augmented accuracy; 
 updating the candidate augmentation transformations list and the default augmentation transformations list in accordance with the affinity metric; 
 removing the second set of augmented training samples from the training data set; 
 augmenting the original training samples and the first set of augmented training samples in accordance with transformations in the updated candidate augmentation transformations list and the updated default augmentation transformations list and storing the augmented training samples as a third set of augmented training samples in the training data set; and 
 training the machine learning model on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, and at least a portion of the third set of augmented training samples to perform the machine learning task. 
 
     
     
       2. The method of  claim 1 , wherein updating the candidate augmentation transformations list and the default augmentation transformations list includes:
 if the affinity metric does not have an acceptable value, updating the candidate augmentation transformations list by removing the selected candidate transformation; 
 if the affinity metric has an acceptable value and the value is less than zero, keeping the selected transformation in the candidate augmentation transformations list; and 
 if the affinity metric has a positive value, removing the selected candidate transformation from the candidate augmentation transformations list and adding the selected candidate transformation to the default augmentation transformations list. 
 
     
     
       3. The method of  claim 1 , wherein applying the default transformation to augment the original training samples includes applying an operation as specified by the default transformation in accordance with operational parameters related to the default transformation and a probability of applying the default transformations. 
     
     
       4. The method of  claim 1 , wherein applying the candidate transformation to augment the original training samples and the first set of augmented training samples includes applying an operation as specified by the candidate transformation in accordance with operational parameters related to the candidate transformation and a probability of applying the candidate transformations. 
     
     
       5. The method of  claim 4  further comprises:
 altering the values of at least one of the operational parameters and the probability if the affinity metric does not have an acceptable value; 
 augmenting the original training samples and the first set of augmented training samples by applying the candidate transformation in accordance with updated operational parameters and the updated probability and storing the augmented training samples as a fourth set of augmented training samples in the training data set; 
 training the machine learning model on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, and at least a portion of the fourth set of augmented training samples; and 
 computing again the affinity metric. 
 
     
     
       6. The method of  claim 1  further comprises traversing individually all candidate transformations in the candidate augmentation transformations list and updating the candidate augmentation transformations list and the default augmentation transformations list accordingly. 
     
     
       7. The method of  claim 1  further comprises selecting n transformations from the updated default augmentation transformations list and the updated candidate augmentation having top n affinity metrics and training the machine learning model in accordance with the original training samples and training samples augmented using n transformations, where n is an integer number. 
     
     
       8. The method of  claim 1 , wherein the training of the machine learning model on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, and at least a portion of the second set of augmented training samples begins once a previous training of the machine learning model on at least a portion of the original training samples and at least a portion of the first set of augmented training samples reaches an accuracy between 50-60%. 
     
     
       9. The method of  claim 1 , wherein the affinity metric is a quantitative difference between the unaugmented accuracy and the augmented accuracy. 
     
     
       10. The method of  claim 1 , wherein the machine learning task comprises at least one of a following tasks:
 an image processing task; 
 a video processing task; 
 an audio processing task; and 
 a text processing task. 
 
     
     
       11. The method of  claim 1 , wherein the machine learning model is previously untrained. 
     
     
       12. The method of  claim 1 , wherein the machine learning model is previously trained on the original training samples. 
     
     
       13. A system for training a machine learning model comprising:
 one or more computers; and 
 one or more storage devices communicatively coupled to the one or more computers, wherein the one or more storage devices store instructions that, when executed by the one or more computers, cause the one or more computers to perform operations comprising:
 receiving a training data set for training the machine learning model to perform a machine learning task, the training data set comprising a plurality of original training samples; 
 augmenting the original training samples by applying default transformations stored in a default augmentation transformations list and storing the augmented training samples as a first set of augmented training samples in the training data set; 
 training the machine learning model on at least a portion of the original training samples and at least a portion of the first set of augmented training samples; 
 computing an unaugmented accuracy of the machine learning model trained on at least a portion of the original training samples and at least a portion of the first set of augmented training samples; 
 augmenting the original training samples and the first set of augmented training samples by applying a candidate transformation selected from a candidate augmentation transformations list and storing the augmented training samples as a second set of augmented training samples in the training data set; 
 training the machine learning model on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, and at least a portion of the second set of augmented training samples; 
 computing an augmented accuracy of the machine learning model trained on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, at least a portion of the second set of augmented training samples; 
 computing an affinity metric from the unaugmented accuracy and the augmented accuracy; 
 updating the candidate augmentation transformations list and the default augmentation transformations list in accordance with the affinity metric; 
 removing the second set of augmented training samples from the training data set; 
 augmenting the original training samples and the first set of augmented training samples in accordance with transformations in the updated candidate augmentation transformations list and the updated default augmentation transformations list and storing the augmented training samples as a third set of augmented training samples in the training data set; and 
 training the machine learning model on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, and at least a portion of the third set of augmented training samples to perform the machine learning task. 
 
 
     
     
       14. The system of  claim 13 , wherein updating the candidate augmentation transformations list and the default augmentation transformations list includes:
 if the affinity metric does not have an acceptable value, updating the candidate augmentation transformations list by removing the selected candidate transformation; 
 if the affinity metric has an acceptable value and the value is less than zero, keeping the selected transformation in the candidate augmentation transformations list; and 
 if the affinity metric has a positive value, removing the selected candidate transformation from the candidate augmentation transformations list and adding the selected candidate transformation to the default augmentation transformations list. 
 
     
     
       15. The system of  claim 13 , wherein applying the default transformation to augment the original training samples includes applying an operation as specified by the default transformation in accordance with operational parameters related to the default transformation and a probability of applying the default transformations. 
     
     
       16. The system of  claim 13 , wherein applying the candidate transformation to augment the original training samples and the first set of augmented training samples includes applying an operation as specified by the candidate transformation in accordance with operational parameters related to the candidate transformation and a probability of applying the candidate transformations. 
     
     
       17. The system of  claim 16  further comprises:
 altering the values of at least one of the operational parameters and the probability if the affinity metric does not have an acceptable value; 
 augmenting the original training samples and the first set of augmented training samples by applying the candidate transformation in accordance with updated operational parameters and the updated probability and storing the augmented training samples as a fourth set of augmented training samples; 
 training the machine learning model on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, and at least a portion of the fourth set of augmented training samples in the training data set; and 
 computing again the affinity metric. 
 
     
     
       18. The system of  claim 13  further comprises traversing individually all candidate transformations in the candidate augmentation transformations list and updating the candidate augmentation transformations list and the default augmentation transformations list accordingly. 
     
     
       19. The system of  claim 13 , wherein the training of the machine learning model on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, and at least a portion of the second set of augmented training samples begins once a previous training of the machine learning model on at least a portion of the original training samples and at least a portion of the first set of augmented training samples reaches an accuracy between 50-60%. 
     
     
       20. One or more non-transitory computer storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations for training a machine learning model comprising:
 receiving a training data set for training the machine learning model to perform a machine learning task, the training data set comprising a plurality of original training samples; 
 augmenting the original training samples by applying default transformations stored in a default augmentation transformations list and storing the augmented training samples as a first set of augmented training samples in the training data set; 
 training the machine learning model on at least a portion of the original training samples and at least a portion of the first set of augmented training samples; 
 computing an unaugmented accuracy of the machine learning model trained on at least a portion of the original training samples and at least a portion of the first set of augmented training samples; 
 augmenting the original training samples and the first set of augmented training samples by applying a candidate transformation selected from a candidate augmentation transformations list and storing the augmented training samples as a second set of augmented training samples in the training data set; 
 training the machine learning model on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, and at least a portion of the second set of augmented training samples; 
 computing an augmented accuracy of the machine learning model trained on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, at least a portion of the second set of augmented training samples; 
 computing an affinity metric from the unaugmented accuracy and the augmented accuracy; 
 updating the candidate augmentation transformations list and the default augmentation transformations list in accordance with the affinity metric; 
 removing the second set of augmented training samples from the training data set; 
 augmenting the original training samples and the first set of augmented training samples in accordance with transformations in the updated candidate augmentation transformations list and the updated default augmentation transformations list and storing the augmented training samples as a third set of augmented training samples in the training data set; and 
 training the machine learning model on at least a portion of the original training samples, at least a portion of the first set of augmented training samples, and at least a portion of the third set of augmented training samples to perform the machine learning task.